src/core/manager.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <assert.h>
  23 #include <errno.h>
  24 #include <string.h>
  25 #include <sys/epoll.h>
  26 #include <signal.h>
  27 #include <sys/signalfd.h>
  28 #include <sys/wait.h>
  29 #include <unistd.h>
  30 #include <sys/poll.h>
  31 #include <sys/reboot.h>
  32 #include <sys/ioctl.h>
  33 #include <linux/kd.h>
  34 #include <termios.h>
  35 #include <fcntl.h>
  36 #include <sys/types.h>
  37 #include <sys/stat.h>
  38 #include <dirent.h>
  39 #include <sys/timerfd.h>
  40
  41 #ifdef HAVE_AUDIT
  42 #include <libaudit.h>
  43 #endif
  44
  45 #include "systemd/sd-daemon.h"
  46 #include "systemd/sd-id128.h"
  47 #include "systemd/sd-messages.h"
  48
  49 #include "manager.h"
  50 #include "transaction.h"
  51 #include "hashmap.h"
  52 #include "macro.h"
  53 #include "strv.h"
  54 #include "log.h"
  55 #include "util.h"
  56 #include "mkdir.h"
  57 #include "ratelimit.h"
  58 #include "cgroup.h"
  59 #include "mount-setup.h"
  60 #include "unit-name.h"
  61 #include "dbus-unit.h"
  62 #include "dbus-job.h"
  63 #include "missing.h"
  64 #include "path-lookup.h"
  65 #include "special.h"
  66 #include "bus-errors.h"
  67 #include "exit-status.h"
  68 #include "virt.h"
  69 #include "watchdog.h"
  70 #include "cgroup-util.h"
  71 #include "path-util.h"
  72 #include "audit-fd.h"
  73 #include "efivars.h"
  74 #include "env-util.h"
  75
  76 /* As soon as 16 units are in our GC queue, make sure to run a gc sweep */
  77 #define GC_QUEUE_ENTRIES_MAX 16
  78
  79 /* As soon as 5s passed since a unit was added to our GC queue, make sure to run a gc sweep */
  80 #define GC_QUEUE_USEC_MAX (10*USEC_PER_SEC)
  81
  82 /* Initial delay and the interval for printing status messages about running jobs */
  83 #define JOBS_IN_PROGRESS_WAIT_SEC 5
  84 #define JOBS_IN_PROGRESS_PERIOD_SEC 1
  85 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
  86
  87 /* Where clients shall send notification messages to */
  88 #define NOTIFY_SOCKET "@/org/freedesktop/systemd1/notify"
  89
  90 #define TIME_T_MAX (time_t)((1UL << ((sizeof(time_t) << 3) - 1)) - 1)
  91
  92 static int manager_setup_notify(Manager *m) {
  93         union {
  94                 struct sockaddr sa;
  95                 struct sockaddr_un un;
  96         } sa;
  97         struct epoll_event ev;
  98         int one = 1, r;
  99
 100         assert(m);
 101
 102         m->notify_watch.type = WATCH_NOTIFY;
 103         m->notify_watch.fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 104         if (m->notify_watch.fd < 0) {
 105                 log_error("Failed to allocate notification socket: %m");
 106                 return -errno;
 107         }
 108
 109         zero(sa);
 110         sa.sa.sa_family = AF_UNIX;
 111
 112         if (getpid() != 1 || detect_container(NULL) > 0)
 113                 snprintf(sa.un.sun_path, sizeof(sa.un.sun_path), NOTIFY_SOCKET "/%llu", random_ull());
 114         else
 115                 strncpy(sa.un.sun_path, NOTIFY_SOCKET, sizeof(sa.un.sun_path));
 116
 117         sa.un.sun_path[0] = 0;
 118
 119         r = bind(m->notify_watch.fd, &sa.sa,
 120                  offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1));
 121         if (r < 0) {
 122                 log_error("bind() failed: %m");
 123                 return -errno;
 124         }
 125
 126         r = setsockopt(m->notify_watch.fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
 127         if (r < 0) {
 128                 log_error("SO_PASSCRED failed: %m");
 129                 return -errno;
 130         }
 131
 132         zero(ev);
 133         ev.events = EPOLLIN;
 134         ev.data.ptr = &m->notify_watch;
 135
 136         r = epoll_ctl(m->epoll_fd, EPOLL_CTL_ADD, m->notify_watch.fd, &ev);
 137         if (r < 0) {
 138                 log_error("Failed to add notification socket fd to epoll: %m");
 139                 return -errno;
 140         }
 141
 142         sa.un.sun_path[0] = '@';
 143         m->notify_socket = strdup(sa.un.sun_path);
 144         if (!m->notify_socket)
 145                 return log_oom();
 146
 147         log_debug("Using notification socket %s", m->notify_socket);
 148
 149         return 0;
 150 }
 151
 152 static int manager_jobs_in_progress_mod_timer(Manager *m) {
 153         struct itimerspec its;
 154
 155         if (m->jobs_in_progress_watch.type != WATCH_JOBS_IN_PROGRESS)
 156                 return 0;
 157
 158         zero(its);
 159
 160         its.it_value.tv_sec = JOBS_IN_PROGRESS_WAIT_SEC;
 161         its.it_interval.tv_sec = JOBS_IN_PROGRESS_PERIOD_SEC;
 162
 163         if (timerfd_settime(m->jobs_in_progress_watch.fd, 0, &its, NULL) < 0)
 164                 return -errno;
 165
 166         return 0;
 167 }
 168
 169 static int manager_watch_jobs_in_progress(Manager *m) {
 170         struct epoll_event ev;
 171         int r;
 172
 173         assert(m);
 174
 175         if (m->jobs_in_progress_watch.type != WATCH_INVALID)
 176                 return 0;
 177
 178         m->jobs_in_progress_watch.type = WATCH_JOBS_IN_PROGRESS;
 179         m->jobs_in_progress_watch.fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC);
 180         if (m->jobs_in_progress_watch.fd < 0) {
 181                 log_error("Failed to create timerfd: %m");
 182                 r = -errno;
 183                 goto err;
 184         }
 185
 186         r = manager_jobs_in_progress_mod_timer(m);
 187         if (r < 0) {
 188                 log_error("Failed to set up timer for jobs progress watch: %s", strerror(-r));
 189                 goto err;
 190         }
 191
 192         zero(ev);
 193         ev.events = EPOLLIN;
 194         ev.data.ptr = &m->jobs_in_progress_watch;
 195
 196         if (epoll_ctl(m->epoll_fd, EPOLL_CTL_ADD, m->jobs_in_progress_watch.fd, &ev) < 0) {
 197                 log_error("Failed to add jobs progress timer fd to epoll: %m");
 198                 r = -errno;
 199                 goto err;
 200         }
 201
 202         log_debug("Set up jobs progress timerfd.");
 203
 204         return 0;
 205
 206 err:
 207         if (m->jobs_in_progress_watch.fd >= 0)
 208                 close_nointr_nofail(m->jobs_in_progress_watch.fd);
 209         watch_init(&m->jobs_in_progress_watch);
 210         return r;
 211 }
 212
 213 static void manager_unwatch_jobs_in_progress(Manager *m) {
 214         if (m->jobs_in_progress_watch.type != WATCH_JOBS_IN_PROGRESS)
 215                 return;
 216
 217         assert_se(epoll_ctl(m->epoll_fd, EPOLL_CTL_DEL, m->jobs_in_progress_watch.fd, NULL) >= 0);
 218         close_nointr_nofail(m->jobs_in_progress_watch.fd);
 219         watch_init(&m->jobs_in_progress_watch);
 220         m->jobs_in_progress_iteration = 0;
 221
 222         log_debug("Closed jobs progress timerfd.");
 223 }
 224
 225 #define CYLON_BUFFER_EXTRA (2*strlen(ANSI_RED_ON) + strlen(ANSI_HIGHLIGHT_RED_ON) + 2*strlen(ANSI_HIGHLIGHT_OFF))
 226 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
 227         char *p = buffer;
 228
 229         assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
 230         assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
 231
 232         if (pos > 1) {
 233                 if (pos > 2)
 234                         p = mempset(p, ' ', pos-2);
 235                 p = stpcpy(p, ANSI_RED_ON);
 236                 *p++ = '*';
 237         }
 238
 239         if (pos > 0 && pos <= width) {
 240                 p = stpcpy(p, ANSI_HIGHLIGHT_RED_ON);
 241                 *p++ = '*';
 242         }
 243
 244         p = stpcpy(p, ANSI_HIGHLIGHT_OFF);
 245
 246         if (pos < width) {
 247                 p = stpcpy(p, ANSI_RED_ON);
 248                 *p++ = '*';
 249                 if (pos < width-1)
 250                         p = mempset(p, ' ', width-1-pos);
 251                 p = stpcpy(p, ANSI_HIGHLIGHT_OFF);
 252         }
 253 }
 254
 255 static void manager_print_jobs_in_progress(Manager *m) {
 256         Iterator i;
 257         Job *j;
 258         char *job_of_n = NULL;
 259         unsigned counter = 0, print_nr;
 260         char cylon[6 + CYLON_BUFFER_EXTRA + 1];
 261         unsigned cylon_pos;
 262
 263         print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
 264
 265         HASHMAP_FOREACH(j, m->jobs, i)
 266                 if (j->state == JOB_RUNNING && counter++ == print_nr)
 267                         break;
 268
 269         /* m->n_running_jobs must be consistent with the contents of m->jobs,
 270          * so the above loop must have succeeded in finding j. */
 271         assert(counter == print_nr + 1);
 272
 273         cylon_pos = m->jobs_in_progress_iteration % 14;
 274         if (cylon_pos >= 8)
 275                 cylon_pos = 14 - cylon_pos;
 276         draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
 277
 278         if (m->n_running_jobs > 1)
 279                 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
 280                         job_of_n = NULL;
 281
 282         manager_status_printf(m, true, cylon, "%sA %s job is running for %s",
 283                               strempty(job_of_n), job_type_to_string(j->type), unit_description(j->unit));
 284         free(job_of_n);
 285
 286         m->jobs_in_progress_iteration++;
 287 }
 288
 289 static int manager_setup_time_change(Manager *m) {
 290         struct epoll_event ev;
 291         struct itimerspec its;
 292
 293         assert(m);
 294         assert(m->time_change_watch.type == WATCH_INVALID);
 295
 296         /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
 297          * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
 298
 299         m->time_change_watch.type = WATCH_TIME_CHANGE;
 300         m->time_change_watch.fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
 301         if (m->time_change_watch.fd < 0) {
 302                 log_error("Failed to create timerfd: %m");
 303                 return -errno;
 304         }
 305
 306         zero(its);
 307
 308         /* We only care for the cancellation event, hence we set the
 309          * timeout to the latest possible value. */
 310         assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
 311         its.it_value.tv_sec = TIME_T_MAX;
 312
 313         if (timerfd_settime(m->time_change_watch.fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
 314                 log_debug("Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
 315                 close_nointr_nofail(m->time_change_watch.fd);
 316                 watch_init(&m->time_change_watch);
 317                 return 0;
 318         }
 319
 320         zero(ev);
 321         ev.events = EPOLLIN;
 322         ev.data.ptr = &m->time_change_watch;
 323
 324         if (epoll_ctl(m->epoll_fd, EPOLL_CTL_ADD, m->time_change_watch.fd, &ev) < 0) {
 325                 log_error("Failed to add timer change fd to epoll: %m");
 326                 return -errno;
 327         }
 328
 329         log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
 330
 331         return 0;
 332 }
 333
 334 static int enable_special_signals(Manager *m) {
 335         int fd;
 336
 337         assert(m);
 338
 339         /* Enable that we get SIGINT on control-alt-del. In containers
 340          * this will fail with EPERM (older) or EINVAL (newer), so
 341          * ignore that. */
 342         if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
 343                 log_warning("Failed to enable ctrl-alt-del handling: %m");
 344
 345         fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
 346         if (fd < 0) {
 347                 /* Support systems without virtual console */
 348                 if (fd != -ENOENT)
 349                         log_warning("Failed to open /dev/tty0: %m");
 350         } else {
 351                 /* Enable that we get SIGWINCH on kbrequest */
 352                 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
 353                         log_warning("Failed to enable kbrequest handling: %s", strerror(errno));
 354
 355                 close_nointr_nofail(fd);
 356         }
 357
 358         return 0;
 359 }
 360
 361 static int manager_setup_signals(Manager *m) {
 362         sigset_t mask;
 363         struct epoll_event ev;
 364         struct sigaction sa;
 365
 366         assert(m);
 367
 368         /* We are not interested in SIGSTOP and friends. */
 369         zero(sa);
 370         sa.sa_handler = SIG_DFL;
 371         sa.sa_flags = SA_NOCLDSTOP|SA_RESTART;
 372         assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 373
 374         assert_se(sigemptyset(&mask) == 0);
 375
 376         sigset_add_many(&mask,
 377                         SIGCHLD,     /* Child died */
 378                         SIGTERM,     /* Reexecute daemon */
 379                         SIGHUP,      /* Reload configuration */
 380                         SIGUSR1,     /* systemd/upstart: reconnect to D-Bus */
 381                         SIGUSR2,     /* systemd: dump status */
 382                         SIGINT,      /* Kernel sends us this on control-alt-del */
 383                         SIGWINCH,    /* Kernel sends us this on kbrequest (alt-arrowup) */
 384                         SIGPWR,      /* Some kernel drivers and upsd send us this on power failure */
 385                         SIGRTMIN+0,  /* systemd: start default.target */
 386                         SIGRTMIN+1,  /* systemd: isolate rescue.target */
 387                         SIGRTMIN+2,  /* systemd: isolate emergency.target */
 388                         SIGRTMIN+3,  /* systemd: start halt.target */
 389                         SIGRTMIN+4,  /* systemd: start poweroff.target */
 390                         SIGRTMIN+5,  /* systemd: start reboot.target */
 391                         SIGRTMIN+6,  /* systemd: start kexec.target */
 392                         SIGRTMIN+13, /* systemd: Immediate halt */
 393                         SIGRTMIN+14, /* systemd: Immediate poweroff */
 394                         SIGRTMIN+15, /* systemd: Immediate reboot */
 395                         SIGRTMIN+16, /* systemd: Immediate kexec */
 396                         SIGRTMIN+20, /* systemd: enable status messages */
 397                         SIGRTMIN+21, /* systemd: disable status messages */
 398                         SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
 399                         SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
 400                         SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
 401                         SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
 402                         SIGRTMIN+27, /* systemd: set log target to console */
 403                         SIGRTMIN+28, /* systemd: set log target to kmsg */
 404                         SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg */
 405                         -1);
 406         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 407
 408         m->signal_watch.type = WATCH_SIGNAL;
 409         m->signal_watch.fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 410         if (m->signal_watch.fd < 0)
 411                 return -errno;
 412
 413         zero(ev);
 414         ev.events = EPOLLIN;
 415         ev.data.ptr = &m->signal_watch;
 416
 417         if (epoll_ctl(m->epoll_fd, EPOLL_CTL_ADD, m->signal_watch.fd, &ev) < 0)
 418                 return -errno;
 419
 420         if (m->running_as == SYSTEMD_SYSTEM)
 421                 return enable_special_signals(m);
 422
 423         return 0;
 424 }
 425
 426 static void manager_strip_environment(Manager *m) {
 427         assert(m);
 428
 429         /* Remove variables from the inherited set that are part of
 430          * the container interface:
 431          * http://www.freedesktop.org/wiki/Software/systemd/ContainerInterface */
 432         strv_remove_prefix(m->environment, "container=");
 433         strv_remove_prefix(m->environment, "container_");
 434
 435         /* Remove variables from the inherited set that are part of
 436          * the initrd interface:
 437          * http://www.freedesktop.org/wiki/Software/systemd/InitrdInterface */
 438         strv_remove_prefix(m->environment, "RD_");
 439
 440         /* Drop invalid entries */
 441         strv_env_clean(m->environment);
 442 }
 443
 444 int manager_new(SystemdRunningAs running_as, Manager **_m) {
 445         Manager *m;
 446         int r = -ENOMEM;
 447
 448         assert(_m);
 449         assert(running_as >= 0);
 450         assert(running_as < _SYSTEMD_RUNNING_AS_MAX);
 451
 452         m = new0(Manager, 1);
 453         if (!m)
 454                 return -ENOMEM;
 455
 456         dual_timestamp_get(&m->userspace_timestamp);
 457         dual_timestamp_from_monotonic(&m->kernel_timestamp, 0);
 458 #ifdef ENABLE_EFI
 459         efi_get_boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
 460 #endif
 461
 462         m->running_as = running_as;
 463         m->name_data_slot = m->conn_data_slot = m->subscribed_data_slot = -1;
 464         m->exit_code = _MANAGER_EXIT_CODE_INVALID;
 465         m->pin_cgroupfs_fd = -1;
 466         m->idle_pipe[0] = m->idle_pipe[1] = -1;
 467
 468         watch_init(&m->signal_watch);
 469         watch_init(&m->mount_watch);
 470         watch_init(&m->swap_watch);
 471         watch_init(&m->udev_watch);
 472         watch_init(&m->time_change_watch);
 473         watch_init(&m->jobs_in_progress_watch);
 474
 475         m->epoll_fd = m->dev_autofs_fd = -1;
 476         m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
 477
 478         m->environment = strv_copy(environ);
 479         if (!m->environment)
 480                 goto fail;
 481
 482         manager_strip_environment(m);
 483
 484         if (running_as == SYSTEMD_SYSTEM) {
 485                 m->default_controllers = strv_new("cpu", NULL);
 486                 if (!m->default_controllers)
 487                         goto fail;
 488         }
 489
 490         if (!(m->units = hashmap_new(string_hash_func, string_compare_func)))
 491                 goto fail;
 492
 493         if (!(m->jobs = hashmap_new(trivial_hash_func, trivial_compare_func)))
 494                 goto fail;
 495
 496         if (!(m->watch_pids = hashmap_new(trivial_hash_func, trivial_compare_func)))
 497                 goto fail;
 498
 499         if (!(m->cgroup_bondings = hashmap_new(string_hash_func, string_compare_func)))
 500                 goto fail;
 501
 502         if (!(m->watch_bus = hashmap_new(string_hash_func, string_compare_func)))
 503                 goto fail;
 504
 505         m->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
 506         if (m->epoll_fd < 0)
 507                 goto fail;
 508
 509         r = manager_setup_signals(m);
 510         if (r < 0)
 511                 goto fail;
 512
 513         r = manager_setup_cgroup(m);
 514         if (r < 0)
 515                 goto fail;
 516
 517         r = manager_setup_notify(m);
 518         if (r < 0)
 519                 goto fail;
 520
 521         r = manager_setup_time_change(m);
 522         if (r < 0)
 523                 goto fail;
 524
 525         /* Try to connect to the busses, if possible. */
 526         r = bus_init(m, running_as != SYSTEMD_SYSTEM);
 527         if (r < 0)
 528                 goto fail;
 529
 530         m->taint_usr = dir_is_empty("/usr") > 0;
 531
 532         *_m = m;
 533         return 0;
 534
 535 fail:
 536         manager_free(m);
 537         return r;
 538 }
 539
 540 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
 541         Unit *u;
 542         unsigned n = 0;
 543
 544         assert(m);
 545
 546         while ((u = m->cleanup_queue)) {
 547                 assert(u->in_cleanup_queue);
 548
 549                 unit_free(u);
 550                 n++;
 551         }
 552
 553         return n;
 554 }
 555
 556 enum {
 557         GC_OFFSET_IN_PATH,  /* This one is on the path we were traveling */
 558         GC_OFFSET_UNSURE,   /* No clue */
 559         GC_OFFSET_GOOD,     /* We still need this unit */
 560         GC_OFFSET_BAD,      /* We don't need this unit anymore */
 561         _GC_OFFSET_MAX
 562 };
 563
 564 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
 565         Iterator i;
 566         Unit *other;
 567         bool is_bad;
 568
 569         assert(u);
 570
 571         if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
 572             u->gc_marker == gc_marker + GC_OFFSET_BAD ||
 573             u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
 574                 return;
 575
 576         if (u->in_cleanup_queue)
 577                 goto bad;
 578
 579         if (unit_check_gc(u))
 580                 goto good;
 581
 582         u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
 583
 584         is_bad = true;
 585
 586         SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
 587                 unit_gc_sweep(other, gc_marker);
 588
 589                 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
 590                         goto good;
 591
 592                 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
 593                         is_bad = false;
 594         }
 595
 596         if (is_bad)
 597                 goto bad;
 598
 599         /* We were unable to find anything out about this entry, so
 600          * let's investigate it later */
 601         u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
 602         unit_add_to_gc_queue(u);
 603         return;
 604
 605 bad:
 606         /* We definitely know that this one is not useful anymore, so
 607          * let's mark it for deletion */
 608         u->gc_marker = gc_marker + GC_OFFSET_BAD;
 609         unit_add_to_cleanup_queue(u);
 610         return;
 611
 612 good:
 613         u->gc_marker = gc_marker + GC_OFFSET_GOOD;
 614 }
 615
 616 static unsigned manager_dispatch_gc_queue(Manager *m) {
 617         Unit *u;
 618         unsigned n = 0;
 619         unsigned gc_marker;
 620
 621         assert(m);
 622
 623         if ((m->n_in_gc_queue < GC_QUEUE_ENTRIES_MAX) &&
 624             (m->gc_queue_timestamp <= 0 ||
 625              (m->gc_queue_timestamp + GC_QUEUE_USEC_MAX) > now(CLOCK_MONOTONIC)))
 626                 return 0;
 627
 628         log_debug("Running GC...");
 629
 630         m->gc_marker += _GC_OFFSET_MAX;
 631         if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
 632                 m->gc_marker = 1;
 633
 634         gc_marker = m->gc_marker;
 635
 636         while ((u = m->gc_queue)) {
 637                 assert(u->in_gc_queue);
 638
 639                 unit_gc_sweep(u, gc_marker);
 640
 641                 LIST_REMOVE(Unit, gc_queue, m->gc_queue, u);
 642                 u->in_gc_queue = false;
 643
 644                 n++;
 645
 646                 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
 647                     u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
 648                         log_debug_unit(u->id, "Collecting %s", u->id);
 649                         u->gc_marker = gc_marker + GC_OFFSET_BAD;
 650                         unit_add_to_cleanup_queue(u);
 651                 }
 652         }
 653
 654         m->n_in_gc_queue = 0;
 655         m->gc_queue_timestamp = 0;
 656
 657         return n;
 658 }
 659
 660 static void manager_clear_jobs_and_units(Manager *m) {
 661         Unit *u;
 662
 663         assert(m);
 664
 665         while ((u = hashmap_first(m->units)))
 666                 unit_free(u);
 667
 668         manager_dispatch_cleanup_queue(m);
 669
 670         assert(!m->load_queue);
 671         assert(!m->run_queue);
 672         assert(!m->dbus_unit_queue);
 673         assert(!m->dbus_job_queue);
 674         assert(!m->cleanup_queue);
 675         assert(!m->gc_queue);
 676
 677         assert(hashmap_isempty(m->jobs));
 678         assert(hashmap_isempty(m->units));
 679
 680         m->n_on_console = 0;
 681         m->n_running_jobs = 0;
 682 }
 683
 684 void manager_free(Manager *m) {
 685         UnitType c;
 686         int i;
 687
 688         assert(m);
 689
 690         manager_clear_jobs_and_units(m);
 691
 692         for (c = 0; c < _UNIT_TYPE_MAX; c++)
 693                 if (unit_vtable[c]->shutdown)
 694                         unit_vtable[c]->shutdown(m);
 695
 696         /* If we reexecute ourselves, we keep the root cgroup
 697          * around */
 698         manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
 699
 700         manager_undo_generators(m);
 701
 702         bus_done(m);
 703
 704         hashmap_free(m->units);
 705         hashmap_free(m->jobs);
 706         hashmap_free(m->watch_pids);
 707         hashmap_free(m->watch_bus);
 708
 709         if (m->epoll_fd >= 0)
 710                 close_nointr_nofail(m->epoll_fd);
 711         if (m->signal_watch.fd >= 0)
 712                 close_nointr_nofail(m->signal_watch.fd);
 713         if (m->notify_watch.fd >= 0)
 714                 close_nointr_nofail(m->notify_watch.fd);
 715         if (m->time_change_watch.fd >= 0)
 716                 close_nointr_nofail(m->time_change_watch.fd);
 717         if (m->jobs_in_progress_watch.fd >= 0)
 718                 close_nointr_nofail(m->jobs_in_progress_watch.fd);
 719
 720         free(m->notify_socket);
 721
 722         lookup_paths_free(&m->lookup_paths);
 723         strv_free(m->environment);
 724
 725         strv_free(m->default_controllers);
 726
 727         hashmap_free(m->cgroup_bondings);
 728         set_free_free(m->unit_path_cache);
 729
 730         close_pipe(m->idle_pipe);
 731
 732         free(m->switch_root);
 733         free(m->switch_root_init);
 734
 735         for (i = 0; i < RLIMIT_NLIMITS; i++)
 736                 free(m->rlimit[i]);
 737
 738         free(m);
 739 }
 740
 741 int manager_enumerate(Manager *m) {
 742         int r = 0, q;
 743         UnitType c;
 744
 745         assert(m);
 746
 747         /* Let's ask every type to load all units from disk/kernel
 748          * that it might know */
 749         for (c = 0; c < _UNIT_TYPE_MAX; c++)
 750                 if (unit_vtable[c]->enumerate)
 751                         if ((q = unit_vtable[c]->enumerate(m)) < 0)
 752                                 r = q;
 753
 754         manager_dispatch_load_queue(m);
 755         return r;
 756 }
 757
 758 int manager_coldplug(Manager *m) {
 759         int r = 0, q;
 760         Iterator i;
 761         Unit *u;
 762         char *k;
 763
 764         assert(m);
 765
 766         /* Then, let's set up their initial state. */
 767         HASHMAP_FOREACH_KEY(u, k, m->units, i) {
 768
 769                 /* ignore aliases */
 770                 if (u->id != k)
 771                         continue;
 772
 773                 if ((q = unit_coldplug(u)) < 0)
 774                         r = q;
 775         }
 776
 777         return r;
 778 }
 779
 780 static void manager_build_unit_path_cache(Manager *m) {
 781         char **i;
 782         DIR _cleanup_free_ *d = NULL;
 783         int r;
 784
 785         assert(m);
 786
 787         set_free_free(m->unit_path_cache);
 788
 789         m->unit_path_cache = set_new(string_hash_func, string_compare_func);
 790         if (!m->unit_path_cache) {
 791                 log_error("Failed to allocate unit path cache.");
 792                 return;
 793         }
 794
 795         /* This simply builds a list of files we know exist, so that
 796          * we don't always have to go to disk */
 797
 798         STRV_FOREACH(i, m->lookup_paths.unit_path) {
 799                 struct dirent *de;
 800
 801                 d = opendir(*i);
 802                 if (!d) {
 803                         if (errno != ENOENT)
 804                                 log_error("Failed to open directory %s: %m", *i);
 805                         continue;
 806                 }
 807
 808                 while ((de = readdir(d))) {
 809                         char *p;
 810
 811                         if (ignore_file(de->d_name))
 812                                 continue;
 813
 814                         p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name, NULL);
 815                         if (!p) {
 816                                 r = -ENOMEM;
 817                                 goto fail;
 818                         }
 819
 820                         r = set_put(m->unit_path_cache, p);
 821                         if (r < 0) {
 822                                 free(p);
 823                                 goto fail;
 824                         }
 825                 }
 826
 827                 closedir(d);
 828                 d = NULL;
 829         }
 830
 831         return;
 832
 833 fail:
 834         log_error("Failed to build unit path cache: %s", strerror(-r));
 835
 836         set_free_free(m->unit_path_cache);
 837         m->unit_path_cache = NULL;
 838 }
 839
 840 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
 841         int r, q;
 842
 843         assert(m);
 844
 845         manager_run_generators(m);
 846
 847         r = lookup_paths_init(
 848                         &m->lookup_paths, m->running_as, true,
 849                         m->generator_unit_path,
 850                         m->generator_unit_path_early,
 851                         m->generator_unit_path_late);
 852         if (r < 0)
 853                 return r;
 854
 855         manager_build_unit_path_cache(m);
 856
 857         /* If we will deserialize make sure that during enumeration
 858          * this is already known, so we increase the counter here
 859          * already */
 860         if (serialization)
 861                 m->n_reloading ++;
 862
 863         /* First, enumerate what we can from all config files */
 864         r = manager_enumerate(m);
 865
 866         /* Second, deserialize if there is something to deserialize */
 867         if (serialization) {
 868                 q = manager_deserialize(m, serialization, fds);
 869                 if (q < 0)
 870                         r = q;
 871         }
 872
 873         /* Any fds left? Find some unit which wants them. This is
 874          * useful to allow container managers to pass some file
 875          * descriptors to us pre-initialized. This enables
 876          * socket-based activation of entire containers. */
 877         if (fdset_size(fds) > 0) {
 878                 q = manager_distribute_fds(m, fds);
 879                 if (q < 0)
 880                         r = q;
 881         }
 882
 883         /* Third, fire things up! */
 884         q = manager_coldplug(m);
 885         if (q < 0)
 886                 r = q;
 887
 888         if (serialization) {
 889                 assert(m->n_reloading > 0);
 890                 m->n_reloading --;
 891         }
 892
 893         return r;
 894 }
 895
 896 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool override, DBusError *e, Job **_ret) {
 897         int r;
 898         Transaction *tr;
 899
 900         assert(m);
 901         assert(type < _JOB_TYPE_MAX);
 902         assert(unit);
 903         assert(mode < _JOB_MODE_MAX);
 904
 905         if (mode == JOB_ISOLATE && type != JOB_START) {
 906                 dbus_set_error(e, BUS_ERROR_INVALID_JOB_MODE, "Isolate is only valid for start.");
 907                 return -EINVAL;
 908         }
 909
 910         if (mode == JOB_ISOLATE && !unit->allow_isolate) {
 911                 dbus_set_error(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
 912                 return -EPERM;
 913         }
 914
 915         log_debug_unit(unit->id,
 916                        "Trying to enqueue job %s/%s/%s", unit->id,
 917                        job_type_to_string(type), job_mode_to_string(mode));
 918
 919         job_type_collapse(&type, unit);
 920
 921         tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
 922         if (!tr)
 923                 return -ENOMEM;
 924
 925         r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, override, false,
 926                                                  mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
 927                                                  mode == JOB_IGNORE_DEPENDENCIES, e);
 928         if (r < 0)
 929                 goto tr_abort;
 930
 931         if (mode == JOB_ISOLATE) {
 932                 r = transaction_add_isolate_jobs(tr, m);
 933                 if (r < 0)
 934                         goto tr_abort;
 935         }
 936
 937         r = transaction_activate(tr, m, mode, e);
 938         if (r < 0)
 939                 goto tr_abort;
 940
 941         log_debug_unit(unit->id,
 942                        "Enqueued job %s/%s as %u", unit->id,
 943                        job_type_to_string(type), (unsigned) tr->anchor_job->id);
 944
 945         if (_ret)
 946                 *_ret = tr->anchor_job;
 947
 948         transaction_free(tr);
 949         return 0;
 950
 951 tr_abort:
 952         transaction_abort(tr);
 953         transaction_free(tr);
 954         return r;
 955 }
 956
 957 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, bool override, DBusError *e, Job **_ret) {
 958         Unit *unit;
 959         int r;
 960
 961         assert(m);
 962         assert(type < _JOB_TYPE_MAX);
 963         assert(name);
 964         assert(mode < _JOB_MODE_MAX);
 965
 966         r = manager_load_unit(m, name, NULL, NULL, &unit);
 967         if (r < 0)
 968                 return r;
 969
 970         return manager_add_job(m, type, unit, mode, override, e, _ret);
 971 }
 972
 973 Job *manager_get_job(Manager *m, uint32_t id) {
 974         assert(m);
 975
 976         return hashmap_get(m->jobs, UINT32_TO_PTR(id));
 977 }
 978
 979 Unit *manager_get_unit(Manager *m, const char *name) {
 980         assert(m);
 981         assert(name);
 982
 983         return hashmap_get(m->units, name);
 984 }
 985
 986 unsigned manager_dispatch_load_queue(Manager *m) {
 987         Unit *u;
 988         unsigned n = 0;
 989
 990         assert(m);
 991
 992         /* Make sure we are not run recursively */
 993         if (m->dispatching_load_queue)
 994                 return 0;
 995
 996         m->dispatching_load_queue = true;
 997
 998         /* Dispatches the load queue. Takes a unit from the queue and
 999          * tries to load its data until the queue is empty */
1000
1001         while ((u = m->load_queue)) {
1002                 assert(u->in_load_queue);
1003
1004                 unit_load(u);
1005                 n++;
1006         }
1007
1008         m->dispatching_load_queue = false;
1009         return n;
1010 }
1011
1012 int manager_load_unit_prepare(Manager *m, const char *name, const char *path, DBusError *e, Unit **_ret) {
1013         Unit *ret;
1014         UnitType t;
1015         int r;
1016
1017         assert(m);
1018         assert(name || path);
1019
1020         /* This will prepare the unit for loading, but not actually
1021          * load anything from disk. */
1022
1023         if (path && !is_path(path)) {
1024                 dbus_set_error(e, BUS_ERROR_INVALID_PATH, "Path %s is not absolute.", path);
1025                 return -EINVAL;
1026         }
1027
1028         if (!name)
1029                 name = path_get_file_name(path);
1030
1031         t = unit_name_to_type(name);
1032
1033         if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, false)) {
1034                 dbus_set_error(e, BUS_ERROR_INVALID_NAME, "Unit name %s is not valid.", name);
1035                 return -EINVAL;
1036         }
1037
1038         ret = manager_get_unit(m, name);
1039         if (ret) {
1040                 *_ret = ret;
1041                 return 1;
1042         }
1043
1044         ret = unit_new(m, unit_vtable[t]->object_size);
1045         if (!ret)
1046                 return -ENOMEM;
1047
1048         if (path) {
1049                 ret->fragment_path = strdup(path);
1050                 if (!ret->fragment_path) {
1051                         unit_free(ret);
1052                         return -ENOMEM;
1053                 }
1054         }
1055
1056         if ((r = unit_add_name(ret, name)) < 0) {
1057                 unit_free(ret);
1058                 return r;
1059         }
1060
1061         unit_add_to_load_queue(ret);
1062         unit_add_to_dbus_queue(ret);
1063         unit_add_to_gc_queue(ret);
1064
1065         if (_ret)
1066                 *_ret = ret;
1067
1068         return 0;
1069 }
1070
1071 int manager_load_unit(Manager *m, const char *name, const char *path, DBusError *e, Unit **_ret) {
1072         int r;
1073
1074         assert(m);
1075
1076         /* This will load the service information files, but not actually
1077          * start any services or anything. */
1078
1079         r = manager_load_unit_prepare(m, name, path, e, _ret);
1080         if (r != 0)
1081                 return r;
1082
1083         manager_dispatch_load_queue(m);
1084
1085         if (_ret)
1086                 *_ret = unit_follow_merge(*_ret);
1087
1088         return 0;
1089 }
1090
1091 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1092         Iterator i;
1093         Job *j;
1094
1095         assert(s);
1096         assert(f);
1097
1098         HASHMAP_FOREACH(j, s->jobs, i)
1099                 job_dump(j, f, prefix);
1100 }
1101
1102 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1103         Iterator i;
1104         Unit *u;
1105         const char *t;
1106
1107         assert(s);
1108         assert(f);
1109
1110         HASHMAP_FOREACH_KEY(u, t, s->units, i)
1111                 if (u->id == t)
1112                         unit_dump(u, f, prefix);
1113 }
1114
1115 void manager_clear_jobs(Manager *m) {
1116         Job *j;
1117
1118         assert(m);
1119
1120         while ((j = hashmap_first(m->jobs)))
1121                 /* No need to recurse. We're cancelling all jobs. */
1122                 job_finish_and_invalidate(j, JOB_CANCELED, false);
1123 }
1124
1125 unsigned manager_dispatch_run_queue(Manager *m) {
1126         Job *j;
1127         unsigned n = 0;
1128
1129         if (m->dispatching_run_queue)
1130                 return 0;
1131
1132         m->dispatching_run_queue = true;
1133
1134         while ((j = m->run_queue)) {
1135                 assert(j->installed);
1136                 assert(j->in_run_queue);
1137
1138                 job_run_and_invalidate(j);
1139                 n++;
1140         }
1141
1142         m->dispatching_run_queue = false;
1143
1144         if (m->n_running_jobs > 0)
1145                 manager_watch_jobs_in_progress(m);
1146
1147         return n;
1148 }
1149
1150 unsigned manager_dispatch_dbus_queue(Manager *m) {
1151         Job *j;
1152         Unit *u;
1153         unsigned n = 0;
1154
1155         assert(m);
1156
1157         if (m->dispatching_dbus_queue)
1158                 return 0;
1159
1160         m->dispatching_dbus_queue = true;
1161
1162         while ((u = m->dbus_unit_queue)) {
1163                 assert(u->in_dbus_queue);
1164
1165                 bus_unit_send_change_signal(u);
1166                 n++;
1167         }
1168
1169         while ((j = m->dbus_job_queue)) {
1170                 assert(j->in_dbus_queue);
1171
1172                 bus_job_send_change_signal(j);
1173                 n++;
1174         }
1175
1176         m->dispatching_dbus_queue = false;
1177         return n;
1178 }
1179
1180 static int manager_process_notify_fd(Manager *m) {
1181         ssize_t n;
1182
1183         assert(m);
1184
1185         for (;;) {
1186                 char buf[4096];
1187                 struct msghdr msghdr;
1188                 struct iovec iovec;
1189                 struct ucred *ucred;
1190                 union {
1191                         struct cmsghdr cmsghdr;
1192                         uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
1193                 } control;
1194                 Unit *u;
1195                 char _cleanup_strv_free_ **tags = NULL;
1196
1197                 zero(iovec);
1198                 iovec.iov_base = buf;
1199                 iovec.iov_len = sizeof(buf)-1;
1200
1201                 zero(control);
1202                 zero(msghdr);
1203                 msghdr.msg_iov = &iovec;
1204                 msghdr.msg_iovlen = 1;
1205                 msghdr.msg_control = &control;
1206                 msghdr.msg_controllen = sizeof(control);
1207
1208                 n = recvmsg(m->notify_watch.fd, &msghdr, MSG_DONTWAIT);
1209                 if (n <= 0) {
1210                         if (n >= 0)
1211                                 return -EIO;
1212
1213                         if (errno == EAGAIN || errno == EINTR)
1214                                 break;
1215
1216                         return -errno;
1217                 }
1218
1219                 if (msghdr.msg_controllen < CMSG_LEN(sizeof(struct ucred)) ||
1220                     control.cmsghdr.cmsg_level != SOL_SOCKET ||
1221                     control.cmsghdr.cmsg_type != SCM_CREDENTIALS ||
1222                     control.cmsghdr.cmsg_len != CMSG_LEN(sizeof(struct ucred))) {
1223                         log_warning("Received notify message without credentials. Ignoring.");
1224                         continue;
1225                 }
1226
1227                 ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr);
1228
1229                 u = hashmap_get(m->watch_pids, LONG_TO_PTR(ucred->pid));
1230                 if (!u) {
1231                         u = cgroup_unit_by_pid(m, ucred->pid);
1232                         if (!u) {
1233                                 log_warning("Cannot find unit for notify message of PID %lu.",
1234                                             (unsigned long) ucred->pid);
1235                                 continue;
1236                         }
1237                 }
1238
1239                 assert((size_t) n < sizeof(buf));
1240                 buf[n] = 0;
1241                 tags = strv_split(buf, "\n\r");
1242                 if (!tags)
1243                         return log_oom();
1244
1245                 log_debug_unit(u->id, "Got notification message for unit %s", u->id);
1246
1247                 if (UNIT_VTABLE(u)->notify_message)
1248                         UNIT_VTABLE(u)->notify_message(u, ucred->pid, tags);
1249         }
1250
1251         return 0;
1252 }
1253
1254 static int manager_dispatch_sigchld(Manager *m) {
1255         assert(m);
1256
1257         for (;;) {
1258                 siginfo_t si;
1259                 Unit *u;
1260                 int r;
1261
1262                 zero(si);
1263
1264                 /* First we call waitd() for a PID and do not reap the
1265                  * zombie. That way we can still access /proc/$PID for
1266                  * it while it is a zombie. */
1267                 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1268
1269                         if (errno == ECHILD)
1270                                 break;
1271
1272                         if (errno == EINTR)
1273                                 continue;
1274
1275                         return -errno;
1276                 }
1277
1278                 if (si.si_pid <= 0)
1279                         break;
1280
1281                 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1282                         char _cleanup_free_ *name = NULL;
1283
1284                         get_process_comm(si.si_pid, &name);
1285                         log_debug("Got SIGCHLD for process %lu (%s)", (unsigned long) si.si_pid, strna(name));
1286                 }
1287
1288                 /* Let's flush any message the dying child might still
1289                  * have queued for us. This ensures that the process
1290                  * still exists in /proc so that we can figure out
1291                  * which cgroup and hence unit it belongs to. */
1292                 r = manager_process_notify_fd(m);
1293                 if (r < 0)
1294                         return r;
1295
1296                 /* And now figure out the unit this belongs to */
1297                 u = hashmap_get(m->watch_pids, LONG_TO_PTR(si.si_pid));
1298                 if (!u)
1299                         u = cgroup_unit_by_pid(m, si.si_pid);
1300
1301                 /* And now, we actually reap the zombie. */
1302                 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1303                         if (errno == EINTR)
1304                                 continue;
1305
1306                         return -errno;
1307                 }
1308
1309                 if (si.si_code != CLD_EXITED && si.si_code != CLD_KILLED && si.si_code != CLD_DUMPED)
1310                         continue;
1311
1312                 log_debug("Child %lu died (code=%s, status=%i/%s)",
1313                           (long unsigned) si.si_pid,
1314                           sigchld_code_to_string(si.si_code),
1315                           si.si_status,
1316                           strna(si.si_code == CLD_EXITED
1317                                 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1318                                 : signal_to_string(si.si_status)));
1319
1320                 if (!u)
1321                         continue;
1322
1323                 log_debug_unit(u->id,
1324                                "Child %lu belongs to %s", (long unsigned) si.si_pid, u->id);
1325
1326                 hashmap_remove(m->watch_pids, LONG_TO_PTR(si.si_pid));
1327                 UNIT_VTABLE(u)->sigchld_event(u, si.si_pid, si.si_code, si.si_status);
1328         }
1329
1330         return 0;
1331 }
1332
1333 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1334         int r;
1335         DBusError error;
1336
1337         dbus_error_init(&error);
1338
1339         log_debug_unit(name, "Activating special unit %s", name);
1340
1341         r = manager_add_job_by_name(m, JOB_START, name, mode, true, &error, NULL);
1342         if (r < 0)
1343                 log_error_unit(name,
1344                                "Failed to enqueue %s job: %s", name, bus_error(&error, r));
1345
1346         dbus_error_free(&error);
1347
1348         return r;
1349 }
1350
1351 static int manager_process_signal_fd(Manager *m) {
1352         ssize_t n;
1353         struct signalfd_siginfo sfsi;
1354         bool sigchld = false;
1355
1356         assert(m);
1357
1358         for (;;) {
1359                 n = read(m->signal_watch.fd, &sfsi, sizeof(sfsi));
1360                 if (n != sizeof(sfsi)) {
1361
1362                         if (n >= 0)
1363                                 return -EIO;
1364
1365                         if (errno == EINTR || errno == EAGAIN)
1366                                 break;
1367
1368                         return -errno;
1369                 }
1370
1371                 if (sfsi.ssi_pid > 0) {
1372                         char *p = NULL;
1373
1374                         get_process_comm(sfsi.ssi_pid, &p);
1375
1376                         log_debug("Received SIG%s from PID %lu (%s).",
1377                                   signal_to_string(sfsi.ssi_signo),
1378                                   (unsigned long) sfsi.ssi_pid, strna(p));
1379                         free(p);
1380                 } else
1381                         log_debug("Received SIG%s.", signal_to_string(sfsi.ssi_signo));
1382
1383                 switch (sfsi.ssi_signo) {
1384
1385                 case SIGCHLD:
1386                         sigchld = true;
1387                         break;
1388
1389                 case SIGTERM:
1390                         if (m->running_as == SYSTEMD_SYSTEM) {
1391                                 /* This is for compatibility with the
1392                                  * original sysvinit */
1393                                 m->exit_code = MANAGER_REEXECUTE;
1394                                 break;
1395                         }
1396
1397                         /* Fall through */
1398
1399                 case SIGINT:
1400                         if (m->running_as == SYSTEMD_SYSTEM) {
1401                                 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE);
1402                                 break;
1403                         }
1404
1405                         /* Run the exit target if there is one, if not, just exit. */
1406                         if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
1407                                 m->exit_code = MANAGER_EXIT;
1408                                 return 0;
1409                         }
1410
1411                         break;
1412
1413                 case SIGWINCH:
1414                         if (m->running_as == SYSTEMD_SYSTEM)
1415                                 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
1416
1417                         /* This is a nop on non-init */
1418                         break;
1419
1420                 case SIGPWR:
1421                         if (m->running_as == SYSTEMD_SYSTEM)
1422                                 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
1423
1424                         /* This is a nop on non-init */
1425                         break;
1426
1427                 case SIGUSR1: {
1428                         Unit *u;
1429
1430                         u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1431
1432                         if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
1433                                 log_info("Trying to reconnect to bus...");
1434                                 bus_init(m, true);
1435                         }
1436
1437                         if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
1438                                 log_info("Loading D-Bus service...");
1439                                 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
1440                         }
1441
1442                         break;
1443                 }
1444
1445                 case SIGUSR2: {
1446                         FILE *f;
1447                         char *dump = NULL;
1448                         size_t size;
1449
1450                         if (!(f = open_memstream(&dump, &size))) {
1451                                 log_warning("Failed to allocate memory stream.");
1452                                 break;
1453                         }
1454
1455                         manager_dump_units(m, f, "\t");
1456                         manager_dump_jobs(m, f, "\t");
1457
1458                         if (ferror(f)) {
1459                                 fclose(f);
1460                                 free(dump);
1461                                 log_warning("Failed to write status stream");
1462                                 break;
1463                         }
1464
1465                         fclose(f);
1466                         log_dump(LOG_INFO, dump);
1467                         free(dump);
1468
1469                         break;
1470                 }
1471
1472                 case SIGHUP:
1473                         m->exit_code = MANAGER_RELOAD;
1474                         break;
1475
1476                 default: {
1477
1478                         /* Starting SIGRTMIN+0 */
1479                         static const char * const target_table[] = {
1480                                 [0] = SPECIAL_DEFAULT_TARGET,
1481                                 [1] = SPECIAL_RESCUE_TARGET,
1482                                 [2] = SPECIAL_EMERGENCY_TARGET,
1483                                 [3] = SPECIAL_HALT_TARGET,
1484                                 [4] = SPECIAL_POWEROFF_TARGET,
1485                                 [5] = SPECIAL_REBOOT_TARGET,
1486                                 [6] = SPECIAL_KEXEC_TARGET
1487                         };
1488
1489                         /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
1490                         static const ManagerExitCode code_table[] = {
1491                                 [0] = MANAGER_HALT,
1492                                 [1] = MANAGER_POWEROFF,
1493                                 [2] = MANAGER_REBOOT,
1494                                 [3] = MANAGER_KEXEC
1495                         };
1496
1497                         if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
1498                             (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
1499                                 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
1500                                 manager_start_target(m, target_table[idx],
1501                                                      (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
1502                                 break;
1503                         }
1504
1505                         if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
1506                             (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
1507                                 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
1508                                 break;
1509                         }
1510
1511                         switch (sfsi.ssi_signo - SIGRTMIN) {
1512
1513                         case 20:
1514                                 log_debug("Enabling showing of status.");
1515                                 manager_set_show_status(m, true);
1516                                 break;
1517
1518                         case 21:
1519                                 log_debug("Disabling showing of status.");
1520                                 manager_set_show_status(m, false);
1521                                 break;
1522
1523                         case 22:
1524                                 log_set_max_level(LOG_DEBUG);
1525                                 log_notice("Setting log level to debug.");
1526                                 break;
1527
1528                         case 23:
1529                                 log_set_max_level(LOG_INFO);
1530                                 log_notice("Setting log level to info.");
1531                                 break;
1532
1533                         case 24:
1534                                 if (m->running_as == SYSTEMD_USER) {
1535                                         m->exit_code = MANAGER_EXIT;
1536                                         return 0;
1537                                 }
1538
1539                                 /* This is a nop on init */
1540                                 break;
1541
1542                         case 26:
1543                                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1544                                 log_notice("Setting log target to journal-or-kmsg.");
1545                                 break;
1546
1547                         case 27:
1548                                 log_set_target(LOG_TARGET_CONSOLE);
1549                                 log_notice("Setting log target to console.");
1550                                 break;
1551
1552                         case 28:
1553                                 log_set_target(LOG_TARGET_KMSG);
1554                                 log_notice("Setting log target to kmsg.");
1555                                 break;
1556
1557                         case 29:
1558                                 log_set_target(LOG_TARGET_SYSLOG_OR_KMSG);
1559                                 log_notice("Setting log target to syslog-or-kmsg.");
1560                                 break;
1561
1562                         default:
1563                                 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
1564                         }
1565                 }
1566                 }
1567         }
1568
1569         if (sigchld)
1570                 return manager_dispatch_sigchld(m);
1571
1572         return 0;
1573 }
1574
1575 static int process_event(Manager *m, struct epoll_event *ev) {
1576         int r;
1577         Watch *w;
1578
1579         assert(m);
1580         assert(ev);
1581
1582         assert_se(w = ev->data.ptr);
1583
1584         if (w->type == WATCH_INVALID)
1585                 return 0;
1586
1587         switch (w->type) {
1588
1589         case WATCH_SIGNAL:
1590
1591                 /* An incoming signal? */
1592                 if (ev->events != EPOLLIN)
1593                         return -EINVAL;
1594
1595                 if ((r = manager_process_signal_fd(m)) < 0)
1596                         return r;
1597
1598                 break;
1599
1600         case WATCH_NOTIFY:
1601
1602                 /* An incoming daemon notification event? */
1603                 if (ev->events != EPOLLIN)
1604                         return -EINVAL;
1605
1606                 if ((r = manager_process_notify_fd(m)) < 0)
1607                         return r;
1608
1609                 break;
1610
1611         case WATCH_FD:
1612
1613                 /* Some fd event, to be dispatched to the units */
1614                 UNIT_VTABLE(w->data.unit)->fd_event(w->data.unit, w->fd, ev->events, w);
1615                 break;
1616
1617         case WATCH_UNIT_TIMER:
1618         case WATCH_JOB_TIMER: {
1619                 uint64_t v;
1620                 ssize_t k;
1621
1622                 /* Some timer event, to be dispatched to the units */
1623                 k = read(w->fd, &v, sizeof(v));
1624                 if (k != sizeof(v)) {
1625
1626                         if (k < 0 && (errno == EINTR || errno == EAGAIN))
1627                                 break;
1628
1629                         log_error("Failed to read timer event counter: %s", k < 0 ? strerror(-k) : "Short read");
1630                         return k < 0 ? -errno : -EIO;
1631                 }
1632
1633                 if (w->type == WATCH_UNIT_TIMER)
1634                         UNIT_VTABLE(w->data.unit)->timer_event(w->data.unit, v, w);
1635                 else
1636                         job_timer_event(w->data.job, v, w);
1637                 break;
1638         }
1639
1640         case WATCH_MOUNT:
1641                 /* Some mount table change, intended for the mount subsystem */
1642                 mount_fd_event(m, ev->events);
1643                 break;
1644
1645         case WATCH_SWAP:
1646                 /* Some swap table change, intended for the swap subsystem */
1647                 swap_fd_event(m, ev->events);
1648                 break;
1649
1650         case WATCH_UDEV:
1651                 /* Some notification from udev, intended for the device subsystem */
1652                 device_fd_event(m, ev->events);
1653                 break;
1654
1655         case WATCH_DBUS_WATCH:
1656                 bus_watch_event(m, w, ev->events);
1657                 break;
1658
1659         case WATCH_DBUS_TIMEOUT:
1660                 bus_timeout_event(m, w, ev->events);
1661                 break;
1662
1663         case WATCH_TIME_CHANGE: {
1664                 Unit *u;
1665                 Iterator i;
1666
1667                 log_struct(LOG_INFO,
1668                            MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
1669                            "MESSAGE=Time has been changed",
1670                            NULL);
1671
1672                 /* Restart the watch */
1673                 epoll_ctl(m->epoll_fd, EPOLL_CTL_DEL, m->time_change_watch.fd,
1674                           NULL);
1675                 close_nointr_nofail(m->time_change_watch.fd);
1676                 watch_init(&m->time_change_watch);
1677                 manager_setup_time_change(m);
1678
1679                 HASHMAP_FOREACH(u, m->units, i) {
1680                         if (UNIT_VTABLE(u)->time_change)
1681                                 UNIT_VTABLE(u)->time_change(u);
1682                 }
1683
1684                 break;
1685         }
1686
1687         case WATCH_JOBS_IN_PROGRESS: {
1688                 uint64_t v;
1689
1690                 /* not interested in the data */
1691                 read(w->fd, &v, sizeof(v));
1692
1693                 manager_print_jobs_in_progress(m);
1694                 break;
1695         }
1696
1697         default:
1698                 log_error("event type=%i", w->type);
1699                 assert_not_reached("Unknown epoll event type.");
1700         }
1701
1702         return 0;
1703 }
1704
1705 int manager_loop(Manager *m) {
1706         int r;
1707
1708         RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
1709
1710         assert(m);
1711         m->exit_code = MANAGER_RUNNING;
1712
1713         /* Release the path cache */
1714         set_free_free(m->unit_path_cache);
1715         m->unit_path_cache = NULL;
1716
1717         manager_check_finished(m);
1718
1719         /* There might still be some zombies hanging around from
1720          * before we were exec()'ed. Leat's reap them */
1721         r = manager_dispatch_sigchld(m);
1722         if (r < 0)
1723                 return r;
1724
1725         while (m->exit_code == MANAGER_RUNNING) {
1726                 struct epoll_event event;
1727                 int n;
1728                 int wait_msec = -1;
1729
1730                 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM)
1731                         watchdog_ping();
1732
1733                 if (!ratelimit_test(&rl)) {
1734                         /* Yay, something is going seriously wrong, pause a little */
1735                         log_warning("Looping too fast. Throttling execution a little.");
1736                         sleep(1);
1737                         continue;
1738                 }
1739
1740                 if (manager_dispatch_load_queue(m) > 0)
1741                         continue;
1742
1743                 if (manager_dispatch_run_queue(m) > 0)
1744                         continue;
1745
1746                 if (bus_dispatch(m) > 0)
1747                         continue;
1748
1749                 if (manager_dispatch_cleanup_queue(m) > 0)
1750                         continue;
1751
1752                 if (manager_dispatch_gc_queue(m) > 0)
1753                         continue;
1754
1755                 if (manager_dispatch_dbus_queue(m) > 0)
1756                         continue;
1757
1758                 if (swap_dispatch_reload(m) > 0)
1759                         continue;
1760
1761                 /* Sleep for half the watchdog time */
1762                 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM) {
1763                         wait_msec = (int) (m->runtime_watchdog / 2 / USEC_PER_MSEC);
1764                         if (wait_msec <= 0)
1765                                 wait_msec = 1;
1766                 } else
1767                         wait_msec = -1;
1768
1769                 n = epoll_wait(m->epoll_fd, &event, 1, wait_msec);
1770                 if (n < 0) {
1771
1772                         if (errno == EINTR)
1773                                 continue;
1774
1775                         return -errno;
1776                 } else if (n == 0)
1777                         continue;
1778
1779                 assert(n == 1);
1780
1781                 r = process_event(m, &event);
1782                 if (r < 0)
1783                         return r;
1784         }
1785
1786         return m->exit_code;
1787 }
1788
1789 int manager_load_unit_from_dbus_path(Manager *m, const char *s, DBusError *e, Unit **_u) {
1790         char *n;
1791         Unit *u;
1792         int r;
1793
1794         assert(m);
1795         assert(s);
1796         assert(_u);
1797
1798         if (!startswith(s, "/org/freedesktop/systemd1/unit/"))
1799                 return -EINVAL;
1800
1801         n = bus_path_unescape(s+31);
1802         if (!n)
1803                 return -ENOMEM;
1804
1805         r = manager_load_unit(m, n, NULL, e, &u);
1806         free(n);
1807
1808         if (r < 0)
1809                 return r;
1810
1811         *_u = u;
1812
1813         return 0;
1814 }
1815
1816 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
1817         Job *j;
1818         unsigned id;
1819         int r;
1820
1821         assert(m);
1822         assert(s);
1823         assert(_j);
1824
1825         if (!startswith(s, "/org/freedesktop/systemd1/job/"))
1826                 return -EINVAL;
1827
1828         r = safe_atou(s + 30, &id);
1829         if (r < 0)
1830                 return r;
1831
1832         j = manager_get_job(m, id);
1833         if (!j)
1834                 return -ENOENT;
1835
1836         *_j = j;
1837
1838         return 0;
1839 }
1840
1841 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
1842
1843 #ifdef HAVE_AUDIT
1844         char *p;
1845         int audit_fd;
1846
1847         audit_fd = get_audit_fd();
1848         if (audit_fd < 0)
1849                 return;
1850
1851         /* Don't generate audit events if the service was already
1852          * started and we're just deserializing */
1853         if (m->n_reloading > 0)
1854                 return;
1855
1856         if (m->running_as != SYSTEMD_SYSTEM)
1857                 return;
1858
1859         if (u->type != UNIT_SERVICE)
1860                 return;
1861
1862         p = unit_name_to_prefix_and_instance(u->id);
1863         if (!p) {
1864                 log_error_unit(u->id,
1865                                "Failed to allocate unit name for audit message: %s", strerror(ENOMEM));
1866                 return;
1867         }
1868
1869         if (audit_log_user_comm_message(audit_fd, type, "", p, NULL, NULL, NULL, success) < 0) {
1870                 if (errno == EPERM) {
1871                         /* We aren't allowed to send audit messages?
1872                          * Then let's not retry again. */
1873                         close_audit_fd();
1874                 } else
1875                         log_warning("Failed to send audit message: %m");
1876         }
1877
1878         free(p);
1879 #endif
1880
1881 }
1882
1883 void manager_send_unit_plymouth(Manager *m, Unit *u) {
1884         int fd = -1;
1885         union sockaddr_union sa;
1886         int n = 0;
1887         char *message = NULL;
1888
1889         /* Don't generate plymouth events if the service was already
1890          * started and we're just deserializing */
1891         if (m->n_reloading > 0)
1892                 return;
1893
1894         if (m->running_as != SYSTEMD_SYSTEM)
1895                 return;
1896
1897         if (u->type != UNIT_SERVICE &&
1898             u->type != UNIT_MOUNT &&
1899             u->type != UNIT_SWAP)
1900                 return;
1901
1902         /* We set SOCK_NONBLOCK here so that we rather drop the
1903          * message then wait for plymouth */
1904         fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1905         if (fd < 0) {
1906                 log_error("socket() failed: %m");
1907                 return;
1908         }
1909
1910         zero(sa);
1911         sa.sa.sa_family = AF_UNIX;
1912         strncpy(sa.un.sun_path+1, "/org/freedesktop/plymouthd", sizeof(sa.un.sun_path)-1);
1913         if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)) < 0) {
1914
1915                 if (errno != EPIPE &&
1916                     errno != EAGAIN &&
1917                     errno != ENOENT &&
1918                     errno != ECONNREFUSED &&
1919                     errno != ECONNRESET &&
1920                     errno != ECONNABORTED)
1921                         log_error("connect() failed: %m");
1922
1923                 goto finish;
1924         }
1925
1926         if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
1927                 log_oom();
1928                 goto finish;
1929         }
1930
1931         errno = 0;
1932         if (write(fd, message, n + 1) != n + 1) {
1933
1934                 if (errno != EPIPE &&
1935                     errno != EAGAIN &&
1936                     errno != ENOENT &&
1937                     errno != ECONNREFUSED &&
1938                     errno != ECONNRESET &&
1939                     errno != ECONNABORTED)
1940                         log_error("Failed to write Plymouth message: %m");
1941
1942                 goto finish;
1943         }
1944
1945 finish:
1946         if (fd >= 0)
1947                 close_nointr_nofail(fd);
1948
1949         free(message);
1950 }
1951
1952 void manager_dispatch_bus_name_owner_changed(
1953                 Manager *m,
1954                 const char *name,
1955                 const char* old_owner,
1956                 const char *new_owner) {
1957
1958         Unit *u;
1959
1960         assert(m);
1961         assert(name);
1962
1963         if (!(u = hashmap_get(m->watch_bus, name)))
1964                 return;
1965
1966         UNIT_VTABLE(u)->bus_name_owner_change(u, name, old_owner, new_owner);
1967 }
1968
1969 void manager_dispatch_bus_query_pid_done(
1970                 Manager *m,
1971                 const char *name,
1972                 pid_t pid) {
1973
1974         Unit *u;
1975
1976         assert(m);
1977         assert(name);
1978         assert(pid >= 1);
1979
1980         if (!(u = hashmap_get(m->watch_bus, name)))
1981                 return;
1982
1983         UNIT_VTABLE(u)->bus_query_pid_done(u, name, pid);
1984 }
1985
1986 int manager_open_serialization(Manager *m, FILE **_f) {
1987         char *path = NULL;
1988         mode_t saved_umask;
1989         int fd;
1990         FILE *f;
1991
1992         assert(_f);
1993
1994         if (m->running_as == SYSTEMD_SYSTEM)
1995                 asprintf(&path, "/run/systemd/dump-%lu-XXXXXX", (unsigned long) getpid());
1996         else
1997                 asprintf(&path, "/tmp/systemd-dump-%lu-XXXXXX", (unsigned long) getpid());
1998
1999         if (!path)
2000                 return -ENOMEM;
2001
2002         saved_umask = umask(0077);
2003         fd = mkostemp(path, O_RDWR|O_CLOEXEC);
2004         umask(saved_umask);
2005
2006         if (fd < 0) {
2007                 free(path);
2008                 return -errno;
2009         }
2010
2011         unlink(path);
2012
2013         log_debug("Serializing state to %s", path);
2014         free(path);
2015
2016         f = fdopen(fd, "w+");
2017         if (!f)
2018                 return -errno;
2019
2020         *_f = f;
2021
2022         return 0;
2023 }
2024
2025 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool serialize_jobs) {
2026         Iterator i;
2027         Unit *u;
2028         const char *t;
2029         char **e;
2030         int r;
2031
2032         assert(m);
2033         assert(f);
2034         assert(fds);
2035
2036         m->n_reloading ++;
2037
2038         fprintf(f, "current-job-id=%i\n", m->current_job_id);
2039         fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2040         fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2041         fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2042
2043         dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2044         dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2045         dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2046         dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2047
2048         if (!in_initrd()) {
2049                 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2050                 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2051         }
2052
2053         STRV_FOREACH(e, m->environment) {
2054                 _cleanup_free_ char *ce;
2055
2056                 ce = cescape(*e);
2057                 if (ce)
2058                         fprintf(f, "env=%s\n", *e);
2059         }
2060
2061         fputc('\n', f);
2062
2063         HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2064                 if (u->id != t)
2065                         continue;
2066
2067                 if (!unit_can_serialize(u))
2068                         continue;
2069
2070                 /* Start marker */
2071                 fputs(u->id, f);
2072                 fputc('\n', f);
2073
2074                 if ((r = unit_serialize(u, f, fds, serialize_jobs)) < 0) {
2075                         m->n_reloading --;
2076                         return r;
2077                 }
2078         }
2079
2080         assert(m->n_reloading > 0);
2081         m->n_reloading --;
2082
2083         if (ferror(f))
2084                 return -EIO;
2085
2086         r = bus_fdset_add_all(m, fds);
2087         if (r < 0)
2088                 return r;
2089
2090         return 0;
2091 }
2092
2093 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2094         int r = 0;
2095
2096         assert(m);
2097         assert(f);
2098
2099         log_debug("Deserializing state...");
2100
2101         m->n_reloading ++;
2102
2103         for (;;) {
2104                 char line[LINE_MAX], *l;
2105
2106                 if (!fgets(line, sizeof(line), f)) {
2107                         if (feof(f))
2108                                 r = 0;
2109                         else
2110                                 r = -errno;
2111
2112                         goto finish;
2113                 }
2114
2115                 char_array_0(line);
2116                 l = strstrip(line);
2117
2118                 if (l[0] == 0)
2119                         break;
2120
2121                 if (startswith(l, "current-job-id=")) {
2122                         uint32_t id;
2123
2124                         if (safe_atou32(l+15, &id) < 0)
2125                                 log_debug("Failed to parse current job id value %s", l+15);
2126                         else
2127                                 m->current_job_id = MAX(m->current_job_id, id);
2128                 } else if (startswith(l, "n-installed-jobs=")) {
2129                         uint32_t n;
2130
2131                         if (safe_atou32(l+17, &n) < 0)
2132                                 log_debug("Failed to parse installed jobs counter %s", l+17);
2133                         else
2134                                 m->n_installed_jobs += n;
2135                 } else if (startswith(l, "n-failed-jobs=")) {
2136                         uint32_t n;
2137
2138                         if (safe_atou32(l+14, &n) < 0)
2139                                 log_debug("Failed to parse failed jobs counter %s", l+14);
2140                         else
2141                                 m->n_failed_jobs += n;
2142                 } else if (startswith(l, "taint-usr=")) {
2143                         int b;
2144
2145                         if ((b = parse_boolean(l+10)) < 0)
2146                                 log_debug("Failed to parse taint /usr flag %s", l+10);
2147                         else
2148                                 m->taint_usr = m->taint_usr || b;
2149                 } else if (startswith(l, "firmware-timestamp="))
2150                         dual_timestamp_deserialize(l+19, &m->firmware_timestamp);
2151                 else if (startswith(l, "loader-timestamp="))
2152                         dual_timestamp_deserialize(l+17, &m->loader_timestamp);
2153                 else if (startswith(l, "kernel-timestamp="))
2154                         dual_timestamp_deserialize(l+17, &m->kernel_timestamp);
2155                 else if (startswith(l, "initrd-timestamp="))
2156                         dual_timestamp_deserialize(l+17, &m->initrd_timestamp);
2157                 else if (startswith(l, "userspace-timestamp="))
2158                         dual_timestamp_deserialize(l+20, &m->userspace_timestamp);
2159                 else if (startswith(l, "finish-timestamp="))
2160                         dual_timestamp_deserialize(l+17, &m->finish_timestamp);
2161                 else if (startswith(l, "env=")) {
2162                         _cleanup_free_ char *uce = NULL;
2163                         char **e;
2164
2165                         uce = cunescape(l+4);
2166                         if (!uce) {
2167                                 r = -ENOMEM;
2168                                 goto finish;
2169                         }
2170
2171                         e = strv_env_set(m->environment, uce);
2172                         if (!e) {
2173                                 r = -ENOMEM;
2174                                 goto finish;
2175                         }
2176
2177                         strv_free(m->environment);
2178                         m->environment = e;
2179                 } else
2180                         log_debug("Unknown serialization item '%s'", l);
2181         }
2182
2183         for (;;) {
2184                 Unit *u;
2185                 char name[UNIT_NAME_MAX+2];
2186
2187                 /* Start marker */
2188                 if (!fgets(name, sizeof(name), f)) {
2189                         if (feof(f))
2190                                 r = 0;
2191                         else
2192                                 r = -errno;
2193
2194                         goto finish;
2195                 }
2196
2197                 char_array_0(name);
2198
2199                 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2200                 if (r < 0)
2201                         goto finish;
2202
2203                 r = unit_deserialize(u, f, fds);
2204                 if (r < 0)
2205                         goto finish;
2206         }
2207
2208 finish:
2209         if (ferror(f)) {
2210                 r = -EIO;
2211                 goto finish;
2212         }
2213
2214         assert(m->n_reloading > 0);
2215         m->n_reloading --;
2216
2217         return r;
2218 }
2219
2220 int manager_distribute_fds(Manager *m, FDSet *fds) {
2221         Unit *u;
2222         Iterator i;
2223         int r;
2224
2225         assert(m);
2226
2227         HASHMAP_FOREACH(u, m->units, i) {
2228
2229                 if (fdset_size(fds) <= 0)
2230                         break;
2231
2232                 if (UNIT_VTABLE(u)->distribute_fds) {
2233                         r = UNIT_VTABLE(u)->distribute_fds(u, fds);
2234                         if (r < 0)
2235                                 return r;
2236                 }
2237         }
2238
2239         return 0;
2240 }
2241
2242 int manager_reload(Manager *m) {
2243         int r, q;
2244         FILE *f;
2245         FDSet *fds;
2246
2247         assert(m);
2248
2249         r = manager_open_serialization(m, &f);
2250         if (r < 0)
2251                 return r;
2252
2253         m->n_reloading ++;
2254
2255         fds = fdset_new();
2256         if (!fds) {
2257                 m->n_reloading --;
2258                 r = -ENOMEM;
2259                 goto finish;
2260         }
2261
2262         r = manager_serialize(m, f, fds, true);
2263         if (r < 0) {
2264                 m->n_reloading --;
2265                 goto finish;
2266         }
2267
2268         if (fseeko(f, 0, SEEK_SET) < 0) {
2269                 m->n_reloading --;
2270                 r = -errno;
2271                 goto finish;
2272         }
2273
2274         /* From here on there is no way back. */
2275         manager_clear_jobs_and_units(m);
2276         manager_undo_generators(m);
2277         lookup_paths_free(&m->lookup_paths);
2278
2279         /* Find new unit paths */
2280         manager_run_generators(m);
2281
2282         q = lookup_paths_init(
2283                         &m->lookup_paths, m->running_as, true,
2284                         m->generator_unit_path,
2285                         m->generator_unit_path_early,
2286                         m->generator_unit_path_late);
2287         if (q < 0)
2288                 r = q;
2289
2290         manager_build_unit_path_cache(m);
2291
2292         /* First, enumerate what we can from all config files */
2293         q = manager_enumerate(m);
2294         if (q < 0)
2295                 r = q;
2296
2297         /* Second, deserialize our stored data */
2298         q = manager_deserialize(m, f, fds);
2299         if (q < 0)
2300                 r = q;
2301
2302         fclose(f);
2303         f = NULL;
2304
2305         /* Third, fire things up! */
2306         q = manager_coldplug(m);
2307         if (q < 0)
2308                 r = q;
2309
2310         assert(m->n_reloading > 0);
2311         m->n_reloading--;
2312
2313 finish:
2314         if (f)
2315                 fclose(f);
2316
2317         if (fds)
2318                 fdset_free(fds);
2319
2320         return r;
2321 }
2322
2323 static bool manager_is_booting_or_shutting_down(Manager *m) {
2324         Unit *u;
2325
2326         assert(m);
2327
2328         /* Is the initial job still around? */
2329         if (manager_get_job(m, m->default_unit_job_id))
2330                 return true;
2331
2332         /* Is there a job for the shutdown target? */
2333         u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
2334         if (u)
2335                 return !!u->job;
2336
2337         return false;
2338 }
2339
2340 bool manager_is_reloading_or_reexecuting(Manager *m) {
2341         assert(m);
2342
2343         return m->n_reloading != 0;
2344 }
2345
2346 void manager_reset_failed(Manager *m) {
2347         Unit *u;
2348         Iterator i;
2349
2350         assert(m);
2351
2352         HASHMAP_FOREACH(u, m->units, i)
2353                 unit_reset_failed(u);
2354 }
2355
2356 bool manager_unit_pending_inactive(Manager *m, const char *name) {
2357         Unit *u;
2358
2359         assert(m);
2360         assert(name);
2361
2362         /* Returns true if the unit is inactive or going down */
2363         u = manager_get_unit(m, name);
2364         if (!u)
2365                 return true;
2366
2367         return unit_pending_inactive(u);
2368 }
2369
2370 void manager_check_finished(Manager *m) {
2371         char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2372         usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2373
2374         assert(m);
2375
2376         if (m->n_running_jobs == 0)
2377                 manager_unwatch_jobs_in_progress(m);
2378
2379         if (hashmap_size(m->jobs) > 0) {
2380                 manager_jobs_in_progress_mod_timer(m);
2381                 return;
2382         }
2383
2384         /* Notify Type=idle units that we are done now */
2385         close_pipe(m->idle_pipe);
2386
2387         /* Turn off confirm spawn now */
2388         m->confirm_spawn = false;
2389
2390         if (dual_timestamp_is_set(&m->finish_timestamp))
2391                 return;
2392
2393         dual_timestamp_get(&m->finish_timestamp);
2394
2395         if (m->running_as == SYSTEMD_SYSTEM && detect_container(NULL) <= 0) {
2396
2397                 /* Note that m->kernel_usec.monotonic is always at 0,
2398                  * and m->firmware_usec.monotonic and
2399                  * m->loader_usec.monotonic should be considered
2400                  * negative values. */
2401
2402                 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2403                 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2404                 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2405                 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2406
2407                 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2408
2409                         kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2410                         initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2411
2412                         if (!log_on_console())
2413                                 log_struct(LOG_INFO,
2414                                            MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2415                                            "KERNEL_USEC=%llu", (unsigned long long) kernel_usec,
2416                                            "INITRD_USEC=%llu", (unsigned long long) initrd_usec,
2417                                            "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec,
2418                                            "MESSAGE=Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2419                                            format_timespan(kernel, sizeof(kernel), kernel_usec),
2420                                            format_timespan(initrd, sizeof(initrd), initrd_usec),
2421                                            format_timespan(userspace, sizeof(userspace), userspace_usec),
2422                                            format_timespan(sum, sizeof(sum), total_usec),
2423                                            NULL);
2424                 } else {
2425                         kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2426                         initrd_usec = 0;
2427
2428                         if (!log_on_console())
2429                                 log_struct(LOG_INFO,
2430                                            MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2431                                            "KERNEL_USEC=%llu", (unsigned long long) kernel_usec,
2432                                            "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec,
2433                                            "MESSAGE=Startup finished in %s (kernel) + %s (userspace) = %s.",
2434                                            format_timespan(kernel, sizeof(kernel), kernel_usec),
2435                                            format_timespan(userspace, sizeof(userspace), userspace_usec),
2436                                            format_timespan(sum, sizeof(sum), total_usec),
2437                                            NULL);
2438                 }
2439         } else {
2440                 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2441                 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2442
2443                 if (!log_on_console())
2444                         log_struct(LOG_INFO,
2445                                    MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2446                                    "USERSPACE_USEC=%llu", (unsigned long long) userspace_usec,
2447                                    "MESSAGE=Startup finished in %s.",
2448                                    format_timespan(sum, sizeof(sum), total_usec),
2449                                    NULL);
2450         }
2451
2452         bus_broadcast_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2453
2454         sd_notifyf(false,
2455                    "READY=1\nSTATUS=Startup finished in %s.",
2456                    format_timespan(sum, sizeof(sum), total_usec));
2457 }
2458
2459 static int create_generator_dir(Manager *m, char **generator, const char *name) {
2460         char *p;
2461         int r;
2462
2463         assert(m);
2464         assert(generator);
2465         assert(name);
2466
2467         if (*generator)
2468                 return 0;
2469
2470         if (m->running_as == SYSTEMD_SYSTEM && getpid() == 1) {
2471
2472                 p = strappend("/run/systemd/", name);
2473                 if (!p)
2474                         return log_oom();
2475
2476                 r = mkdir_p_label(p, 0755);
2477                 if (r < 0) {
2478                         log_error("Failed to create generator directory %s: %s",
2479                                   p, strerror(-r));
2480                         free(p);
2481                         return r;
2482                 }
2483         } else {
2484                 p = strjoin("/tmp/systemd-", name, ".XXXXXX", NULL);
2485                 if (!p)
2486                         return log_oom();
2487
2488                 if (!mkdtemp(p)) {
2489                         log_error("Failed to create generator directory %s: %m",
2490                                   p);
2491                         free(p);
2492                         return -errno;
2493                 }
2494         }
2495
2496         *generator = p;
2497         return 0;
2498 }
2499
2500 static void trim_generator_dir(Manager *m, char **generator) {
2501         assert(m);
2502         assert(generator);
2503
2504         if (!*generator)
2505                 return;
2506
2507         if (rmdir(*generator) >= 0) {
2508                 free(*generator);
2509                 *generator = NULL;
2510         }
2511
2512         return;
2513 }
2514
2515 void manager_run_generators(Manager *m) {
2516         DIR *d = NULL;
2517         const char *generator_path;
2518         const char *argv[5];
2519         mode_t u;
2520         int r;
2521
2522         assert(m);
2523
2524         generator_path = m->running_as == SYSTEMD_SYSTEM ? SYSTEM_GENERATOR_PATH : USER_GENERATOR_PATH;
2525         d = opendir(generator_path);
2526         if (!d) {
2527                 if (errno == ENOENT)
2528                         return;
2529
2530                 log_error("Failed to enumerate generator directory %s: %m",
2531                           generator_path);
2532                 return;
2533         }
2534
2535         r = create_generator_dir(m, &m->generator_unit_path, "generator");
2536         if (r < 0)
2537                 goto finish;
2538
2539         r = create_generator_dir(m, &m->generator_unit_path_early, "generator.early");
2540         if (r < 0)
2541                 goto finish;
2542
2543         r = create_generator_dir(m, &m->generator_unit_path_late, "generator.late");
2544         if (r < 0)
2545                 goto finish;
2546
2547         argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
2548         argv[1] = m->generator_unit_path;
2549         argv[2] = m->generator_unit_path_early;
2550         argv[3] = m->generator_unit_path_late;
2551         argv[4] = NULL;
2552
2553         u = umask(0022);
2554         execute_directory(generator_path, d, (char**) argv);
2555         umask(u);
2556
2557         trim_generator_dir(m, &m->generator_unit_path);
2558         trim_generator_dir(m, &m->generator_unit_path_early);
2559         trim_generator_dir(m, &m->generator_unit_path_late);
2560
2561 finish:
2562         if (d)
2563                 closedir(d);
2564 }
2565
2566 static void remove_generator_dir(Manager *m, char **generator) {
2567         assert(m);
2568         assert(generator);
2569
2570         if (!*generator)
2571                 return;
2572
2573         strv_remove(m->lookup_paths.unit_path, *generator);
2574         rm_rf(*generator, false, true, false);
2575
2576         free(*generator);
2577         *generator = NULL;
2578 }
2579
2580 void manager_undo_generators(Manager *m) {
2581         assert(m);
2582
2583         remove_generator_dir(m, &m->generator_unit_path);
2584         remove_generator_dir(m, &m->generator_unit_path_early);
2585         remove_generator_dir(m, &m->generator_unit_path_late);
2586 }
2587
2588 int manager_set_default_controllers(Manager *m, char **controllers) {
2589         char **l;
2590
2591         assert(m);
2592
2593         l = strv_copy(controllers);
2594         if (!l)
2595                 return -ENOMEM;
2596
2597         strv_free(m->default_controllers);
2598         m->default_controllers = l;
2599
2600         cg_shorten_controllers(m->default_controllers);
2601
2602         return 0;
2603 }
2604
2605 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
2606         int i;
2607
2608         assert(m);
2609
2610         for (i = 0; i < RLIMIT_NLIMITS; i++) {
2611                 if (!default_rlimit[i])
2612                         continue;
2613
2614                 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
2615                 if (!m->rlimit[i])
2616                         return -ENOMEM;
2617         }
2618
2619         return 0;
2620 }
2621
2622 void manager_recheck_journal(Manager *m) {
2623         Unit *u;
2624
2625         assert(m);
2626
2627         if (m->running_as != SYSTEMD_SYSTEM)
2628                 return;
2629
2630         u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
2631         if (u && SOCKET(u)->state != SOCKET_RUNNING) {
2632                 log_close_journal();
2633                 return;
2634         }
2635
2636         u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
2637         if (u && SERVICE(u)->state != SERVICE_RUNNING) {
2638                 log_close_journal();
2639                 return;
2640         }
2641
2642         /* Hmm, OK, so the socket is fully up and the service is up
2643          * too, then let's make use of the thing. */
2644         log_open();
2645 }
2646
2647 void manager_set_show_status(Manager *m, bool b) {
2648         assert(m);
2649
2650         if (m->running_as != SYSTEMD_SYSTEM)
2651                 return;
2652
2653         m->show_status = b;
2654
2655         if (b)
2656                 touch("/run/systemd/show-status");
2657         else
2658                 unlink("/run/systemd/show-status");
2659 }
2660
2661 static bool manager_get_show_status(Manager *m) {
2662         assert(m);
2663
2664         if (m->running_as != SYSTEMD_SYSTEM)
2665                 return false;
2666
2667         if (m->show_status)
2668                 return true;
2669
2670         /* If Plymouth is running make sure we show the status, so
2671          * that there's something nice to see when people press Esc */
2672
2673         return plymouth_running();
2674 }
2675
2676 void manager_status_printf(Manager *m, bool ephemeral, const char *status, const char *format, ...) {
2677         va_list ap;
2678
2679         if (!manager_get_show_status(m))
2680                 return;
2681
2682         /* XXX We should totally drop the check for ephemeral here
2683          * and thus effectively make 'Type=idle' pointless. */
2684         if (ephemeral && m->n_on_console > 0)
2685                 return;
2686
2687         if (!manager_is_booting_or_shutting_down(m))
2688                 return;
2689
2690         va_start(ap, format);
2691         status_vprintf(status, true, ephemeral, format, ap);
2692         va_end(ap);
2693 }
2694
2695 void watch_init(Watch *w) {
2696         assert(w);
2697
2698         w->type = WATCH_INVALID;
2699         w->fd = -1;
2700 }